Skip to content

Quickstart

This takes you from an installed ProdPilot to your editor’s agent auditing and fixing a project. It assumes you have installed it.

Terminal window
prodpilot setup

It asks for the GitHub token and the Render API key, writes them to ~/.prodpilot/config.toml, restricts the file to your account and checks that it did. Nothing you type is printed back. Real output:

ProdPilot credential setup
Credentials are stored in E:\tmp\dev\.prodpilot\config.toml
They are never written into a project directory.
GitHub token: Render API key:
Saved to E:\tmp\dev\.prodpilot\config.toml
Permissions: access is limited to the current user

This capture set PRODPILOT_CONFIG_HOME to a scratch folder so no real credentials were involved; normally the path is ~/.prodpilot/config.toml. Run it again at any time, and press enter to keep a stored value.

You only need the credentials to deploy. Auditing and fixing work without them.

Terminal window
prodpilot doctor

Real output on the machine this site was built on:

ProdPilot doctor
Config file: E:\tmp\dev\.prodpilot\config.toml
found
github_token: stored
render_api_key: stored
permissions: access is limited to the current user
Tools ProdPilot runs:
git: git version 2.54.0.windows.1
Node.js: v26.3.0
Docker: daemon 29.7.2
All checks passed.

Doctor never prints a credential, only whether it is stored. If anything is missing it names it and exits non-zero, so a script can see the failure.

Terminal window
prodpilot connect vscode
prodpilot connect cursor
prodpilot connect devin --project PATH

Each writes the editor’s own configuration in the place that editor documents, naming the prodpilot command on this machine by its absolute path. VS Code and Cursor are connected once for every project; Windsurf, now Devin, once per project. Each editor has its own page: VS Code, Cursor, Windsurf (now Devin).

Open your project and ask the agent:

Use ProdPilot to audit this project and fix what it reports, one rule at a time, reporting each fix with prodpilot_fix_applied.

The agent calls prodpilot_detect_stack, then asks prodpilot_fix_instruction for each failing rule, applies what it says and reports back with prodpilot_fix_applied. ProdPilot answers resolved, unresolved or blocked from the rule’s own checker, whatever the agent claimed. The MCP tools reference shows real responses from each tool.

On one of the deliberately broken sample projects in the ProdPilot repository, node_express_insecure, the first audit scores 13 out of 100 with 6 critical failures, and after the fix loop it scores 96 with none. That run is recorded in full.

Commit your own changes first, then ask:

Deploy this project with prodpilot_deploy.

The deploy creates a real Render service on the free plan and pushes to your GitHub repository, so the tool tells the agent to confirm with you before calling it. It runs the scoring gate and then the eight ProdPush stages, and stops at the first that fails, saying which and why. On success you get a live URL and a GitHub Actions workflow that redeploys on every push.